>There's still a way around it. If people keep it loaded, the program can still track their message submissions, or display requests.
>
>Based on their login date, and the time of day, the program could surmise that they are using it several times per day. That would be an implied multiple login for the day, rather a one-time, hour or less, login period.
>
>Plus, if there is one login that lasts more than a day, and there is still activity on the account, then it could mark it as 'everyday'. Otherwise, the next login will start the whole process again. There is no need for interactivity in this case.

Granted, and these are points I thought about after my rather failed attempt to track usage, but by then you're getting into some more serious work, and it just wasn't all that important :)

Plus, you're doing some subjective guesswork doing it that programmatic way, which technically is not a very good sampling method. (Not that the popup method is necessarily very good methodology, either, you must rely on honesty, at least). And there wasn't any exact answer for me, either - some days I'm in and out many times, sometimes I go for a few days with one or no logins at all...so my choice was only approximate.

I guess it all depends on the objective of the study what technique makes the most sense, and you must have all possible answers available, or a comment section, maybe. Not an easy subject to get good data results.

Polling is kind of trendy right now - Delphi forums (in other subjects) have a new polling feature that's quite popular (but optional to participate).